copywriteR CNV identification

Question

[gambrdella]

Hi All, I am using copywriteR to identify somatic copy number variants in a matched experiment in which I have normal and tumor. I runned copywriteR just fine following the R vignette. At the end of the process after I have called plotCNA function I can load the results that are in the segment.Rdata object.

Here if I understood well in the column seg.mean have for each segment the Log2ratio value of each segment. Now, my question is, how I can define what is amplified or deleted? If I understood well reading around, doing 2^log2ratio should give me number of copies. But is it possible to assign a pvalue for each segment that tells how much that segment should be altered? Or how can I selected a threshold in log2ratio to define what is altered and what not?

Thank you in advance for any help and clarification.

> head(segment.CNA.object$output)
ID chrom  loc.start   loc.end num.mark seg.mean
1 log2.tum.bam.vs.log2.normal.bam     1   825000.5  41575000      787  -0.1760
2 log2.tum.bam.vs.log2.normal.bam     1 41625000.5  93175000     1016   0.3496
3 log2.tum.bam.vs.log2.normal.bam     1 93225000.5  93275000        2   2.3676
4 log2.tum.bam.vs.log2.normal.bam     1 93325000.5  93475000        4   0.7452
5 log2.tum.bam.vs.log2.normal.bam     1 93525000.5  94275000       16   0.1227
6 log2.tum.bam.vs.log2.normal.bam     1 94325000.5 121575000      505  -0.1780

Answer 1

[ReWeeda]

Well, 2^(Log2Ratio) doesn't give the number of copy. Reversing the logarithm gives you the original ratio between sample and control.

Taking the first line of the output you provided, reversing the log2ratio you can find that the sample has got a number of copies in that region that's about 0.89 * the number of copy the control has got in the same region.

Given that ratio = 1 means no differences in the number of copies, and that log2(1)=0, more your value is distant from 0 greater the difference between the number of reads is.

I'm not sure if copywriteR filters out less significant regions before providing you an output (I suggest to check the manual) but you can plot the distribution of your values and decide to keep only those regions whose ratio is significantly higher or lower in the sample given the control.

[gambrdella]

thanks for the answer. It is what I am actually doing plotting the distribution of these values and selecting significant altered regions from there. However I would have know if there was something more standard from who usually work with that kind of data.

[ReWeeda]

I can suggest you to take into consideration the number of reads mapped to a segment if you have enough samples. Using dummy numbers: Suppose that the number of mapped reads in segment A spans between 10 and 30 for normal and tumor while the number of mapped reads for segment B spans between 1000 and 1500. Segment A will results in a fold change of 3 while segment B will results in a fold change of 1.5 but segment B is far more significant.